Centrality determination of Pb-Pb collisions at $\sqrt{s_{\rm NN}}$ = 2.76 TeV with ALICE

This publication describes the methods used to measure the centrality of inelastic Pb-Pb collisions at a center-of-mass energy of 2.76 TeV per colliding nucleon pair with ALICE. The centrality is a key parameter in the study of the properties of QCD matter at extreme temperature and energy density, because it is directly related to the initial overlap region of the colliding nuclei. Geometrical properties of the collision, such as the number of participating nucleons and number of binary nucleon-nucleon collisions, are deduced from a Glauber model with a sharp impact parameter selection, and shown to be consistent with those extracted from the data. The centrality determination provides a tool to compare ALICE measurements with those of other experiments and with theoretical calculations.


Introduction
Ultra-relativistic heavy-ion collisions at the Large Hadron Collider (LHC) produce strongly interacting matter under extreme conditions of temperature and energy density, similar to those prevailing in the first few microseconds after the Big Bang [1].
Since nuclei are extended objects, the volume of the interacting region depends on the impact parameter (b) of the collision, defined as the distance between the centers of the two colliding nuclei in a plane transverse to the beam axis.It is customary in the field of heavy-ion physics to introduce the concept of the centrality of the collision, which is directly related to the impact parameter, and inferred by comparison of data with simulations of the collisions.
The purely geometrical Glauber model [2], which typically is used in this context, has its origins in the quantum mechanical model for p-A and A-A scattering described in [3][4][5].The model treats a nuclear collision as a superposition of binary nucleon-nucleon interactions.The volume of the initial overlap region is expressed via the number of participant nucleons.A participant nucleon of one nucleus is defined as a nucleon that undergoes one or more binary collisions with nucleons of the other nucleus.The number of participants and spectators, N part and N spec = 2A − N part , where A is the total number of nucleons in the nucleus (mass number), and the number of binary collisions N coll are calculated for a given value of the impact parameter and for a realistic initial distribution of nucleons inside the nucleus, and assuming that nucleons follow straight trajectories.This approach provides a consistent description of p-A, d-A, and A-A collisions, and is especially useful when comparing data from different experiments or from different collision systems and to theoretical calculations.
Neither the impact parameter nor geometrical quantities, such as N part , N spec , or N coll are directly measurable.Two experimental observables related to the collision geometry are the average charged-particle multiplicity N ch and the energy carried by particles close to the beam direction and deposited in Zero-Degree Calorimeters (ZDC), called the zero-degree energy E ZDC .The average charged-particle multiplicity is assumed to decrease monotonically with increasing impact parameter.The energy deposited in the zero-degree calorimeters, E ZDC , is directly related to the number of spectator nucleons N spec , which constitute the part of the nuclear volume not involved in the interaction.However, unlike N ch , E ZDC does not depend monotonically on the impact parameter b because nucleons bound in nuclear fragments with similar magnetic rigidity as the beam nuclei remain inside the beam-pipe and therefore are not detected in the ZDC.Since fragment formation is more important in peripheral collisions, the monotonic relationship between E ZDC and b is valid only for relatively central events (small b).For this reason, the zero-degree energy measurement needs to be combined with another observable that is monotonically correlated with b.
The centrality is usually expressed as a percentage of the total nuclear interaction cross section σ [2].The centrality percentile c of an A-A collision with an impact parameter b is defined by integrating the impact parameter distribution dσ /db as In ALICE, the centrality is defined as the percentile of the hadronic cross section corresponding to a particle multiplicity above a given threshold (N The procedure can be simplified by replacing the cross section with the number of observed events, corrected for the trigger efficiency.However, at LHC energies, the strong electromagnetic fields generated by the heavy ions moving at relativistic velocity lead to large cross sections for QED processes [6][7][8][9].Although the cross sections for these processes exceed those for the hadronic cross section by several orders of magnitude, they only contaminate the hadronic cross section in the most peripheral collisions.For this reason one may choose to restrict the centrality determination to the region where such contamination is negligible.The fraction of hadronic events excluded by such cut as well as the trigger efficiency can be estimated using a model of the nuclear collision and the related particle production.
In this paper, we report on the centrality determination used in the analyses of the Pb-Pb collision data from the 2010 and 2011 run recorded with the ALICE detector [10].Specifically the analysis presented here is done with a subset of the 2010 data, but the methods and results are valid for 2011 as well.In Section 2, we describe the implementation of the Glauber model used by ALICE.We extract mean numbers of the relevant geometrical quantities for typical centrality classes defined by classifying the events according to their impact parameter.Section 3 describes the experimental conditions and the event selection with particular emphasis on the rejection of QED and machine-induced backgrounds.Section 4 presents the methods employed by ALICE for the determination of the hadronic cross section, needed for the absolute determination of the centrality.The main method uses the VZERO amplitude distribution fitted with the Glauber model.The systematic uncertainty is obtained by comparing the fit to an unbiased VZERO distribution obtained by correcting the measured one by the efficiency of the event selection and the purity of the event sample.Section 5 presents the determination of the centrality classes using either the multiplicity at mid-rapidity or the energy deposited in the ZDC.We discuss the relation between the measured multiplicity and geometrical quantities connected to centrality, established by the Glauber Model.These are nearly identical to those obtained in Section 2, classifying the events according to their impact parameter, which are therefore used as reference in all ALICE analyses.Section 6 presents the precision of the centrality determination in ALICE.Section 7 summarizes and concludes the paper.

The Glauber Model
The Glauber model is widely used to describe the dependence of N part and N coll on b in p-A, d-A and A-A collisions [2][3][4][5].The purpose of Monte Carlo implementations of the Glauber model [19,20] is to compose two nuclei out of nucleons and simulate their collision process eventby-event.Geometrical quantities are calculated by simulating many nucleus-nucleus collisions.Mean values of these quantities are calculated for centrality classes defined by classifying the events according to their impact parameter b.
Following [21], the first step in the Glauber Monte Carlo is to prepare a model of the two nuclei by defining stochastically the position of the nucleons in each nucleus.The nucleon position in the 208 Pb nucleus is determined by the nuclear density function, modeled by the functional  [11][12][13].The σ el NN curve is a fit performed by the COMPETE Collaboration also available at [12,14].The pp data from ATLAS [15], CMS [16], TOTEM, [17] and ALICE [18] agree well with the interpolation for σ inel NN .
form (modified Woods-Saxon or 2-parameter Fermi distribution): The parameters are based on data from low energy electron-nucleus scattering experiments [22].
Protons and neutrons are assumed to have the same nuclear profile.The parameter ρ 0 is the nucleon density, which provides the overall normalization, not relevant for the Monte Carlo simulation, R = (6.62 ± 0.06) fm is the radius parameter of the 208 Pb nucleus and a = (0.546 ± 0.010) fm is the skin thickness of the nucleus, which indicates how quickly the nuclear density falls off near the edge of the nucleus.The additional parameter w is needed to describe nuclei whose maximum density is reached at radii r > 0 (w = 0 for Pb).In the Monte Carlo procedure the radial coordinate of a nucleon is randomly drawn from the distribution 4πr 2 ρ(r) and ρ 0 is determined by the overall normalization condition ρ(r)d 3 r = A. We require a hard-sphere exclusion distance of d min = 0.4 fm between the centers of the nucleons, i.e. no pair of nucleons inside the nucleus has a distance less than d min .The hard-sphere exclusion distance, characteristic of the length of the repulsive nucleon-nucleon force, is not known experimentally and thus is varied by 100% (d min = (0.4 ± 0.4) fm).
The second step is to simulate a nuclear collision.The impact parameter b is randomly selected from the geometrical distribution dP/db ∼ b up to a maximum b max 20 fm > 2R Pb .The maximum value of the impact parameter b max is chosen large enough to simulate collisions until the interaction probability becomes zero.This is particularly important for the calculation of the total Pb-Pb cross section.The nucleus-nucleus collision is treated as a sequence of independent binary nucleon-nucleon collisions, where the nucleons travel on straight-line trajectories and the inelastic nucleon-nucleon cross section is assumed to be independent of the number of collisions a nucleon underwent previously, i.e. the same cross section is used for all successive collisions.Two nucleons from different nuclei are assumed to collide if the relative transverse distance between centers is less than the distance corresponding to the inelastic nucleon-nucleon cross section d < σ inel NN /π.A Gaussian overlap function can be used as an alternative to the blackdisk nucleon-nucleon overlap function [23].It makes no significant difference within systematic uncertainty in the global event properties.The number of collisions N coll and the number of participants N part are determined by counting, respectively, the binary nucleon collisions and the nucleons that experience at least one collision.Following the notation in [2], the geometric nuclear overlap function T AA is then calculated as T AA = N coll /σ inel NN , and represents the effective nucleon luminosity in the collision process.
For nuclear collisions at √ s NN = 2.76 TeV, we use σ inel NN = (64 ± 5) mb, estimated by interpolation [11] of pp data at different center-of-mass energies and from cosmic rays [12,14], and subtracting the elastic scattering cross section from the total cross section.The interpolation is in good agreement with the ALICE measurement of the pp inelastic cross section at √ s NN = 2.76 TeV, σ inel NN = (62.8± 2.4 +1.2 −4.0 ) mb [18], and with the measurements of ATLAS [15], CMS [16], and TOTEM [17] at √ s NN = 7 TeV, as shown in Fig. 1.
The total Pb-Pb cross section is calculated as σ PbPb = N evt (N coll ≥ 1)/N evt (N coll ≥ 0) × πb 2 max , i.e. the geometrical value corrected by the fraction of events with at least one nucleon-nucleon collision.We obtain σ PbPb = (7.64 ± 0.22(syst.))b, in agreement with the ALICE measurement σ PbPb = (7.7 ± 0.1(stat.)+0.6 −0.5 (syst.))b [9].Table 1 report the mean number of participants N part and collisions N coll , and the mean nuclear thickness function T AA for centrality classes defined by sharp cuts in the impact parameter b calculated with the Glauber model (Fig. 2).The root mean square (RMS) of these distributions is a measure for the magnitude of the dispersion of the quantities.
The systematic uncertainties on the mean values are obtained by independently varying the parameters of the Glauber model within their estimated uncertainties.More specifically, the default value of the nucleon-nucleon cross section of σ inel NN = 64 mb was varied between 59 mb and 69 mb.The Woods-Saxon parameters were varied by one standard deviation to determine uncertainties related to the nuclear density profile.The minimum distance of 0.4 fm between two nucleons of the same nucleus was varied by 100%, from 0 to 0.8 fm to evaluate the effects of a nucleon hard core (as mentioned above).Figure 3 shows the resulting variations for Pb-Pb collisions at √ s NN = 2.76 TeV.The total systematic uncertainty reported in Table 1 was obtained by adding in quadrature the deviations from the default result for each of the variations listed above.The uncertainty of N part ranges from about 3-4% in peripheral collisions to < 1% in central collisions, the uncertainty of N coll ranges from about 7% in peripheral collisions to about 11% in central collisions, the uncertainty of T AA ranges from about 6% in peripheral collisions to about 3% in central collisions.The nuclear overlap function T AA is often used to compare observables related to hard processes in A-A and pp collisions.Since T AA = N coll /σ inel NN , it has the same systematic uncertainties as N coll except that the uncertainty on σ inel NN cancels out.Finally it is worth noting that more sophisticated implementations of the Glauber model [23][24][25] suggest that effects not included in our Glauber model, such as the changes of the excluded volume on the nuclear density and two-body correlations, can be approximated by slightly adjusting the Woods-Saxon parameters.The modified parameters, however, are well covered by the systematic uncertainty quoted above for the parameters that we use.ALICE is an experiment dedicated to the study of heavy-ion collisions at the LHC.A detailed description of the apparatus is given in Ref. [10].Here, we briefly describe the detector components used in this analysis.
The Silicon Pixel Detector (SPD) is the innermost part of the Inner Tracking System (ITS).It consists of two cylindrical layers of hybrid silicon pixel assemblies positioned at average radial distances of 3.9 and 7.6 cm from the beam line, with a total of 9.8 × 10 6 pixels of size 50 × 425 µm 2 , read out by 1200 electronic chips.The SPD coverage for particles originating from the center of the detector is |η| < 2.0 and |η| < 1.4 for the inner and outer layers, respectively.Each chip provides a fast signal if at least one of its pixels is hit.The signals from the 1200 chips are combined in a programmable logic unit which supplies a trigger signal.The fraction of SPD channels active during 2010 data taking was 70% for the inner and 78% for the outer layers.
The VZERO detector consists of two arrays of 32 scintillator cells placed at distances z = 3.4 m and z = −0.9m from the nominal interaction point, along the beam line, covering the full azimuth.The VZERO detector is within 2.8 < η < 5.1 (VZERO-A) and −3.7 < η < −1.7 (VZERO-C).Both amplitude and time of signals in each scintillator are recorded.The VZERO time resolution is better than 1 ns, allowing discrimination of beam-beam collisions from background events produced upstream of the experiment.The VZERO is also used to provide a trigger signal (see 3.2).
The Time Projection Chamber (TPC) is used for charged particle trajectory reconstruction, track momentum measurement and particle identification.The ALICE TPC is a large cylindrical drift detector whose active volume extends radially from 85 to 247 cm, and from -250 to +250 cm along the beam direction.The active volume of nearly 90 m 3 is filled with a gas mixture of Ne (85.7%),CO 2 (9.5%) and N 2 (4.8%) until the end of 2010, and Ne (90%) and CO 2 (10%) since the beginning of 2011.A central electrode maintained at -100 kV divides the TPC into two sections.The end-caps are equipped with multiwire proportional chambers with cathode pad readout.For a particle traversing the TPC up to 159 position signals (clusters) are recorded.The cluster data are used to reconstruct the charged particle trajectory as well as to calculate the particle's specific energy loss used to identify the species of the particle which has produced the track.
The two Zero Degree Calorimeters (ZDC) in the ALICE experiment measure the energy of spectator (non-interacting) nucleons: ZP measures protons and ZN measures neutrons.They are situated about 114 m from the interaction point on each side of the experiment [10].Each ZDC consists of two quartz fiber sampling calorimeters: the neutron calorimeter positioned between the two beam pipes downstream of the first machine dipole that separates the two charged particle beams, and the proton calorimeter positioned externally to the outgoing beam pipe.The energy resolution at beam energy is estimated to be 20% for the neutron (20.0% for ZNC, 21.2% for ZNA) and 24% for the proton calorimeters, respectively.

Data set and online event selection
During the first LHC Pb-Pb run in 2010, beams of four bunches with about 10 7 Pb ions per bunch collided at √ s NN = 2.76 TeV, with an estimated luminosity of 5 × 10 23 cm −2 s −1 .AL-ICE collected about 90 million nuclear collision events using different interaction triggers with increasingly tighter conditions.These triggers used VZERO and SPD detector signals in coincidence with a bunch crossing corresponding to a beam-beam collision: The threshold in the VZERO detector for each of the VZERO tiles corresponded approximately to the energy deposition of one minimum ionizing particle.
Control events were also collected with the same trigger logic, in coincidence with only one beam crossing the ALICE interaction point (from either the A or the C side) or with no beam at all ("empty").The luminous region had an RMS width of 5.9 cm in the longitudinal direction and 50 µm in the transverse direction.For the estimated luminosity, using the least selective of the interaction triggers, the observed rate was about 50 Hz.This was mainly due to electromagnetically induced processes [26].These processes have large cross sections at LHC energies but generate low multiplicities and therefore do not contribute to the typical particle multiplicities of interest for the present paper.The trigger rate without beam was negligible and the rate in coincidence with bunches of only one beam was about 1 Hz.The probability for collision pile-up per triggered event was less than 10 −4 .
Fig. 5: (Color online) Correlation between the sum and the difference of times recorded by the neutron ZDC on either side of the interaction region.The large cluster in the middle corresponds to collisions between ions in the nominal RF buckets of each beam, while the small clusters along the diagonals (spaced by 2.5 ns in the time difference) correspond to collisions in which one of the ions is displaced by one or more RF buckets.

Offline event selection
The offline event selection is applied with the purpose of selecting hadronic interactions with the highest possible efficiency, while rejecting the machine-induced and physical backgrounds.The offline event selection replays the on-line trigger condition, using the same quantities calculated offline, so that events triggered by noise in the SPD are discarded, and the weighted time average over all channels is used for the VZERO, leading to a better time resolution.In addition, the offline event selection rejects the machine-induced background and parasitic collisions.This contamination amounts to about 25% of all collected events.To keep the conditions of all detectors as uniform as possible (in particular those around mid-rapidity, such as the SPD), the centrality analysis was restricted to a region around the vertex, |z vtx | 10 cm.

Machine-Induced Background
One source of machine-induced background is due to beam-gas events, caused by one of the beams interacting with the residual gas in the beam-pipe; another source of background are events where ions in the beam halo interact with mechanical structures in the machine.These interactions mostly occur outside of the interaction region and thus produce a signal that is "too early" in the same-side VZERO, compared to a collision that occurs in the nominal interaction region between the VZERO detectors.Therefore these events can be rejected using the timing information of the VZERO.This is illustrated in Fig. 4 which shows the arrival time of particles at the VZERO A detector relative to the nominal beam crossing time.Beam-halo or beam-gas interactions are visible as secondary peaks in the time distribution because particles produced in background interactions arrive at earlier times in the detector relative to particles produced in beam-beam collisions at the nominal vertex, which are the majority of the signals.Other small peaks between these main ones arise from satellite collisions.
Another source of machine-induced background is due to parasitic collisions from debunched ions.The radio-frequency (RF) structure of the LHC of 400 MHz is such that there are 10 equidistant RF buckets within the 25 ns time interval between two possible nominal bunch positions.Therefore the buckets are spaced by 2.5 ns.Only one of them should be populated by ions [27].However, ions can "jump" into one of the neighboring buckets.Therefore collisions occur either between ions in the nominal RF buckets but also between one or two ions displaced by one or more RF buckets.This causes a displacement in the Z-vertex position of 2.5 ns/2 • c = 37.5 cm, well outside the fiducial region |z vtx | 10 cm.Those events are thus to be considered as "background" and are rejected using the correlation between the sum and the difference of times measured in each of the neutron ZDCs, as shown in Fig. 5.Such satellite collisions can also be rejected using the vertex cut.
After the event selection, the remaining machine-induced background, estimated from the control triggers (i.e.triggers that fire for coincidences between empty and filled or empty and empty bunches), is negligible.

Electromagnetic interaction background
At the LHC energy, the cross sections for electromagnetic (EM) processes, generated by the EM fields of relativistic heavy ions, are enormous (O(kbarn)) [6][7][8][9].This is the main physical background, and needs to be rejected in heavy-ion collisions to isolate hadronic interactions.QED processes consist of: photo-production and photo-nuclear interactions.Photo-production results in the creation of an e + e − pair.Photo-nuclear interactions, where one photon from the EM field of one of the nuclei interacts with the other nucleus, possibly fluctuating to a vector meson, yield a low multiplicity of soft particles in the ALICE central barrel.In the case of single photo-production the particle multiplicity is asymmetric within the event.Along the beam direction the electromagnetic dissociation (EMD) cross sections are large resulting in a non-negligible probability for one neutron emission from either nucleus.
The EMD cross sections have been measured in a special run triggering on a signal in one of the neutron ZDCs, ZNA or ZNC, with a threshold placed well below the single neutron signal to detect the neutrons from Giant Dipole Resonance (GDR) decay emitted very close to beam rapidity [9].The recorded event sample is dominated by electromagnetic dissociation of one or both nuclei measured to be σ single EMD = 187.4± 0.2(stat.)+13.2 −11.2 (syst.)b compared to the mutual EMD cross section of σ mutual EMD = 5.7 ± 0.1(stat.)± 0.4(syst.)b.The single EMD events can be clearly identified when correlating the response of ZNA and ZNCs (Fig. 6).The additional requirement of a signal in an electromagnetic calorimeter close to beam rapidity (ZEM) allows one to distinguish between mutual EMD and hadronic interaction events.In order to reduce the contribution due to single neutron emission, we require a ZDC signal three standard deviations above the single neutron peak.This selection rejects about 3% of all events within 10 cm from the nominal interaction point after removal of beam-gas and parasitic collisions, and only removes events for peripheral collisions (in the 90-100% region).The coincidence of the ZDC signals rejects all the single neutron emission events.The simultanous emissions from both nuclei still are accepted, which, however, are only relevant for very peripheral collisions.
For systematic studies, another selection based on the information from the TPC is used, where at least one track reconstructed in the TPC is requested in order to keep the event.This selection removes few peripheral hadronic interactions, and strongly suppresses the EM background.

Determination of the hadronic cross section
In order to classify the collisions in percentiles of the hadronic cross section using the charged particle multiplicity, it is necessary to know the particle multiplicity at which the purity of the event sample and the efficiency of the event selection becomes 100%.We define the Anchor Point (AP) as the amplitude of the VZERO detector equivalent to 90% of the hadronic cross section, which determines the absolute scale of the centrality.The determination of the AP requires the knowledge of the trigger efficiency and the remaining background contamination in nuclear collision events.Two methods have been used to study this.The difference in the results obtained with the two methods is used to estimate the systematic uncertainty by: − Simulating the multiplicity distribution (see Sec. 4.1).In the first approach, we use a full simulation of hadronic and EM processes, including a detailed description of the detector response, to study the efficiency of the event selection (Sec.4.1.1)and to estimate the background contamination (Sec.4.1.2).The real multiplicity distribution, corrected for efficiency and purity, allows direct access to the AP.
− Fitting the multiplicity distribution (see Sec. 4.2).In the second method, we use the Glauber Monte Carlo, combined with a simple model for particle production, to simulate a multiplicity distribution which is then compared to the experimental one.The simulated distribution describes the experimental one down to the most peripheral events where they start to deviate due to background contamination and limited trigger efficiency.The location of the divergence between the data and simulation can be used to define the AP.
The centrality determination is performed for different trigger and detector settings.The different triggers change the fraction of accepted events from EM processes, which determines the shape of the multiplicity distribution for very peripheral collisions below the AP.The position of the AP is very stable for the entire 2010 (and 2011) run period and does not change within the quoted systematic uncertainty discussed below.Small variations in detector conditions induce small changes in the position of the edge of the multiplicity distribution for most central events.Nevertheless, the centrality determination, adjusted to account for small changes in the detector configuration, provides a stable centrality selection for the entire data-taking period (the mean fraction of events in the 0-1% bin is 0.01 with a RMS of 0.001, and in the 40-50% bin the mean fraction is 0.101 with an RMS of 0.002).

Method 1: Correcting the multiplicity distribution
With this method the AP is determined by evaluating the efficiency of the event selection and by estimating the purity of the obtained event sample.

Efficiency of the event selection
The efficiency for the different event selections is studied with simulations of hadronic reactions and with dedicated pp runs.For simulations we use HIJING [28] or AMPT [29] with a full GEANT [30] description of the ALICE detector and a trigger emulator.In the simulations the efficiency is defined as the ratio of events selected by a given condition to all generated events.In the two dedicated pp runs, which were taken at the end of the 2010 run at √ s = 7 TeV, the detector conditions were similar to those in the Pb-Pb run.The VZERO gain was adjusted such that the response to minimum ionizing particles (MIP) corresponded to 3 ADC channels for one run, or 6 for the other run.For the Pb-Pb run, it was set to 4 channels, i.e. between the two tested conditions.In the special pp runs, we used a minimum interaction trigger, which requires a logical OR between a hit in the SPD and in either of the two VZERO detectors (CINT1 trigger condition).The same event selection criteria as used in the Pb-Pb run as the trigger have been applied.The relative event selection efficiency is defined as the ratio of events selected by a given condition to all the events recorded with the pp minimum-bias interaction trigger (CINT1).Since the pp minimum-bias interaction trigger (CINT1) has an efficiency that is effectively 100% for non-diffractive events [18], the relative efficiency measured in the pp runs, shown in Fig. 7, can be qualitatively compared to that obtained in Pb-Pb simulations with N part = 2. Except for very low amplitudes, results from HIJING and AMPT are in very good agreement.AMPT predicts a slightly higher efficiency (about 0.5%), as a consequence of the broader rapidity distribution.The comparison with the pp runs shows a reasonable agreement for the "MIP = 6" case, while the "MIP = 3" is clearly lower.
For the Pb-Pb run, the efficiency of the event selection is calculated using the average of results obtained with HIJING and AMPT.The efficiency of the interaction triggers is 99.4%, 97.1%, 96.9% respectively for 2-out-of-3, V0AND, 3-out-of-3 using HIJING and 99.7%, 98.6%, 98.4% using AMPT.The line in Fig. 7, corresponding to the 90% of the hadronic cross section, shows that the trigger is always fully efficient for the 90% most central collisions, except for the "MIP = 3" pp case, where the efficiency is 95%.

Remaining contamination
The purity of the data sample passing a given event selection is estimated using HIJING simulations [28] for hadronic processes and QED [8] and STARLIGHT [7] for the simulations of the EM background.For the electromagnetic dissociation we assume that the selection based on the signal 3σ above the single-neutron peak in the ZDCs (see 3.3.2) is fully efficient.
In Fig. 8, data taken with the V0AND interaction trigger are compared to the sum of HIJING and background (QED + STARLIGHT) simulations with the same event selection.The simulations are scaled to the known cross sections: − HIJING (hadronic): σ H = 7.66 b [28]; [8]; The sum of the simulations is normalized to the data in the region 150 < VZERO amplitude < 500, where there is no background contamination.The contribution from QED is completely removed by the V0AND trigger.The dashed lines, indicating respectively 80% and 90% of the hadronic cross section, show that there is no significant background contamination for collisions more central than 90%.The region 90-100% is reasonably understood as the agreement between data and simulation is quite good.The remaining discrepancy between the data and the sum of all contributions is included in the systematic uncertainty.
To assign a systematic uncertainty, the comparison is made for the three online interaction triggers and other event selections requiring (i) V0AND + TPC: one track fully reconstructed in the TPC on top of the V0AND trigger; (ii) V0AND + ZDC: 3σ cut above single neutron peak in ZDC on top of the V0AND trigger.For all these variations a cross section is calculated and the difference is included in the systematic uncertainty.
Figure 9 shows the purity of the various Pb-Pb event samples after those selections.The purity, plotted as a function of the VZERO amplitude (V), is defined as the fraction of hadronic collisions over all the events selected with a given condition: where σ x and N x are the cross sections and number of events for a given process, x, where x = H, SNS, SND, and Q, for HIJING, STARLIGHT single, STARLIGHT double, and QED, respectively.
The purity of the event sample can be verified using the correlation of the energy deposition in the two sides of the ZN calorimeter, similar to the one shown in Fig. 6.Single neutron peaks are visible in the 80-90% centrality class, which may indicate some remaining contamination from EMD events.However their origin can be also attributed to asymmetric Pb-Pb events, as well as a pile up of an EMD and a hadronic collision.Since this contamination can not be easily removed, analyses that use peripheral classes like 80-90% assign an additional 6% systematic uncertainty on the event selection to take into account the possible contamination from EMD.

Method 2: Fitting the multiplicity distribution
Another independent way to define the AP uses a phenomenological approach based on the Glauber Monte Carlo to fit the experimental multiplicity distribution.The Glauber Monte Carlo uses the assumptions mentioned above plus a convolution of a model for particle production, based on a negative binomial distribution (NBD).This latter assumption is motivated by the fact that in minimum bias pp and pp collisions at high energy, the charged particle multiplicity dσ /dN ch has been measured over a wide range of rapidity and is well described by a NBD [31,32].This approach allows one to simulate an experimental multiplicity distribution (e.g.VZERO amplitude), which can be compared with the one from data.
Figure 10 shows the distribution of VZERO amplitudes for all events triggered with the 3-out-of-3 trigger (see 3.2) after removing the beam background (see 3.  N part and dN ch /dη and to detector acceptance and resolution.
The Glauber Monte Carlo defines, for an event with a given impact parameter b, the corresponding N part and N coll .The particle multiplicity per nucleon-nucleon collision is parametrized by a NBD.To apply this model to any collision with a given N part and N coll value we introduce the concept of "ancestors", i.e. independently emitting sources of particles.We assume that the number of ancestors N ancestors can be parameterized by This is inspired by two-component models [33,34], which decompose nucleus-nucleus collisions into soft and hard interactions, where the soft interactions produce particles with an average multiplicity proportional to N part , and the probability for hard interactions to occur is proportional to N coll .We discuss the independence of the fit results of this assumption below (Sec.4.2.1).
To generate the number of particles produced per interaction, we use the negative binomial distribution which gives the probability of measuring n hits per ancestor, where µ is the mean multiplicity per ancestor and k controls the width.For every Glauber Monte Carlo event, the NBD is sampled N ancestors times to obtain the averaged simulated VZERO amplitude for this event, which is proportional to the number of particles hitting the hodoscopes.The VZERO amplitude distribution is simulated for an ensemble of events and for various values of the NBD parameters µ, k, and the N ancestors parameter f .A minimization procedure is applied to find the parameters which result in the smallest χ 2 , also shown in Fig. 10.The fit is performed for VZERO amplitudes large enough so that the purity of the event sample and the efficiency of the event selection is 100%.That leaves a very broad range in the amplitude values that can be fitted to extract parameters f , µ and k directly from the data.The amplitude, above which we have 90% of the hadronic cross section, defines the AP.The quality of the fit is good, as the χ 2 /NDF is approximately unity for all fits.We note that the high multiplicity tail, which is quite sensitive to fluctuations and the detector resolution not implemented in the model, is not perfectly well described.Even replacing the black-disk nucleon-nucleon overlap function with a Gaussian does not improve the fit, as the difference in the N part distribution is washed out in the N ch distribution.However, it is important to remark that the fit is used solely to determine the AP, which is quite insensitive to the detailed shape of the high multiplicity tail.
An equivalent procedure was applied to fit, with the NBD-Glauber method, the distribution of the hits collected in the outer layer of the SPD, and the tracks reconstructed in the TPC.All these analyses give consistent results, which are summarized in Sec.4.3.

Ancestor dependence
The number of emitting sources N ancestors is determined by a function inspired by the twocomponent models, i.e.N ancestors = f • N part + (1 − f ) • N coll .However, other assumptions can be made leading to a different parametrization, which are briefly discussed in the following.The ancestor dependence on N part and N coll derives from a parametrization of the dependence of the charged particle multiplicity on N part and N coll .Systematic studies of this dependence performed at the SPS [36][37][38], at RHIC [39], and recently at the LHC [35,40,41], have been used in an attempt to constrain different models of particle production.The charged particle multiplicity is expected to scale with N part in scenarios dominated by soft processes.In this case, all the participant nucleons can be assumed to contribute with the same amount of energy to particle production, and the scaling with N part is approximately linear.By contrast, a scaling with N coll is expected for nuclear collisions in an energy regime where hard processes dominate over soft particle production.In this case, nuclear collisions can be considered as a superposition of binary nucleon-nucleon collisions.Two-component models are used to quantify the relative importance of soft and hard processes in the particle production mechanism at different energies.
To determine the scaling behavior of the particle production, the charged particle multiplicity dN ch /dη as a function of the number of participants N part was fitted with a power-law function of N part i.e. dN ch /dη ∝ N α part .While at SPS energy the scaling with N part is approximately linear, i.e. α ∼ 1 [36][37][38], results from the experiments at RHIC show evidence of a large contribution of hard processes to particle production, resulting in α > 1.
The charged particle multiplicity per participant pair dN ch /dη / 0.5N part measured by AL-ICE [35] is fitted (Fig. 11) with three different parametrizations of the ancestor dependence mentioned above: and the fit parameters are reported in Tab. 2. We note that the value obtained for f is in a good agreement with the value obtained in the NBD-Glauber fit, shown in Fig. 10.
While the value obtained for α and for β with the power-law parametrization of N part and N coll indicate that neither of these scalings perfectly describes the data (α > 1 and β < 1), we note that the value of α is similar to that measured at RHIC (1.16 ± 0.04 [39]) and slightly higher than that at the SPS (α ∼ 1, see [36] for a review).The results obtained with the twocomponent model, where 0 < f < 1, indicate that both the contribution of N part and N coll are needed to explain the particle production confirm this.However, the χ 2 /NDF reported in Table 2, indicate an equally good fit for all models, thus revealing that no unique physics conclusion can be drawn from such fits and that the particular choice of parametrization has no influence on the results of the centrality determination.

Systematic uncertainty on the Anchor Point
The determination of the AP by either correcting or fitting the multiplicity distribution is evaluated in Fig. 12 by comparing the VZERO amplitude distributions for various event selections.The systematic uncertainty on the AP is estimated by comparing the percentage of the hadronic cross section at the VZERO amplitude chosen as the AP (V0 AP ) obtained correcting or fitting the multiplicity distribution.For the first method (Sec.4.1), we used the results from the HIJING and AMPT simulations.For the second method (Sec.4.2), we used alternative centrality definitions based on (i) TPC tracks; (ii) SPD hit multiplicities, and obtaining a value for the V0 AP using the correlation between SPD or TPC and VZERO; (iii) different ranges for the Glauber model fit; (iv) different ancestor dependence of the particle production model to a power law of N part ; (v) different nucleon-nucleon cross section and parameters of the Woods-Saxon distribution within their estimated uncertainties.All the results, compared in Table 3, allow to define the AP as the VZERO amplitude above which we obtain 90% of the hadronic cross section with the NBD-Glauber fit (the baseline in Tab.3) with a systematic uncertainty of 1%, determined as the RMS of all the results presented in Table 3.The variations of the AP are not part of the quoted systematic uncertainties for N part , N coll and T AA , which include only variations of the Glauber parameters.The uncertainty on the AP is typically included in our analyses as an uncertainty on the limits of the centrality classes and propagated into an uncertainty on the specific measured observable.
5 Centrality classes and their relation with geometrical quantities

Determination of the centrality classes with the multiplicity distributions
The percentile of the hadronic cross section is determined for any value of the VZERO amplitude by integrating the measured VZERO amplitude distribution normalized at the anchor point V0 AP , i.e. 90% of the hadronic cross section.For example, if we define V as the VZERO amplitude, the top 10% central class is defined by the boundary V0 10 which satisfies The same is done for the number of clusters in the SPD and the number of reconstructed tracks in the TPC.The events with multiplicity lower than that of the anchor point, contaminated by EM background and trigger inefficiency, are not used in the physics analyses.
One can divide the experimental distribution into classes, by defining sharp cuts on e.g.VZERO amplitude, which correspond to well defined percentile intervals of the hadronic cross section.The number of centrality classes that one can define is connected with the resolution achieved on the quantities used in the definition.In general, centrality classes are defined so that the separation between the central values of b and N part for two adjacent classes is significantly larger than the resolution of that variable (see Sec. 6).

Finding the number of participants with the multiplicity distributions
In Sec.4.2 we fit the measured VZERO amplitude distribution with the amplitude distribution simulated with the NBD-Glauber.This creates a connection between an experimental observable and the geometrical model of nuclear collisions used in the Glauber Monte Carlo.From this we can access the geometrical properties, like N part , N coll , T AA .A given centrality class, defined by sharp cuts in the measured distribution, corresponds to the same class in the simulated distribution.For the simulated distribution we retain the input information from the Glauber model.Therefore, we can calculate the mean number of participants N part , the mean number of collisions N coll and the average nuclear overlap function T AA for centrality classes defined by sharp cuts in the simulated multiplicity distribution, corresponding to given percentiles of the hadronic cross section.As shown in Table A.1, the mean values and their dispersions differ from those calculated for geometrical classes, defined by sharp cuts in the impact parameter b (Table 1), by less than 1% for the most central classes (up to about 50%) and by less than 2% for the most peripheral ones (above 50%).This confirms that multiplicity fluctuations and detector resolution only play a minor role in the centrality determination.

Determination of the centrality classes with the ZDC
Another way to determine the centrality is to measure the energy deposited by the spectators in the ZDC.The spectator neutrons and protons having a rapidity close to that of the beam, are detected in the ZDC.Naively, measurement of the number of spectator neutrons and protons would give direct measurement of the number of participants since N part would simply be given by where E ZDC is the energy measured in the ZDC, A = 208 is the mass number of Pb, and E A is the beam energy per nucleon.However, fragment formation amongst the spectator nucleons breaks the simple linear and monotonic relation in the measured variables, since some spectator nucleons are bound into light nuclear fragments that have a charge over mass ratio similar to the beam, therefore, remaining inside the beam-pipe and are undetected by the ZDC [42,43].This effect becomes quantitatively important for peripheral events and therefore Eq. 7 cannot be used as a reliable estimate of N part .
Consequently, the ZDC information needs to be correlated to another quantity that has a monotonic relation with N part .In our case, we use the energy measured by two small EM calorimeters (ZEM).These detectors are placed only on the A side about 7.5 m from the interaction point, covering the region 4.8 < η < 5.7 [10].
Since the ZDC calorimeters are far from the interaction region, and therefore have an acceptance insensitive to the vertex position, a centrality measurement based on the ZDC is particularly suited for any analysis that does not require a vertex cut [44].
Centrality classes are defined by cuts on the two-dimensional distribution of the ZDC energy as a function of the ZEM amplitude.The ZDC signal is proportional to N part for central events, while the ZEM amplitude is an unknown function of N part and N coll .Therefore the definition of the centrality classes in this two-dimensional space is not trivial.As shown in Fig. 13, centrality classes are defined by using the centrality classes defined previously with the VZERO amplitude to determine regions in the ZDC-ZEM plane, corresponding to a given centrality.The boundaries between centrality classes, or the points belonging to the same narrow centrality class c (c ± δ c) can be fitted with linear functions.All these lines are found to intersect at a common point.Using this common point, we refitted the boundaries of the various centrality classes with the linear functions shown in Fig. 13.
As can be seen from the figure, the slopes of the fitted functions increase going from central to peripheral collisions and tend to infinity, as the lines become almost straight vertical lines, when approaching the point where the correlation between ZDC and ZEM inverts its sign.The value of the slope that defines a centrality class in the ZDC vs ZEM phase-space is proportional to the tangent of the percentile, which implies that the percentiles behave like an angle in the ZDC vs ZEM phase-space.
This function of ZDC and ZEM can then be used as centrality estimator for the most central events (0-30%) above the turning-point of ZDC. Figure 14 shows the distribution of the VZERO amplitude for all triggered events and for various centrality classes selected with this method.

Resolution of the centrality determination
As described above, two independent methods are used to determine experimentally the centrality of the collision.The first one uses the multiplicity distributions from various detectors covering different pseudo-rapidity ranges.Specifically we use the sum of the amplitude in the VZERO detectors (A and C side), the number of clusters in the outer layer of the SPD detector, and the number of tracks reconstructed in the TPC.The second method uses the ZDC correlated with the ZEM.
The accuracy of the experimental determination of the centrality was evaluated by comparing the different estimates event-by-event.For example, in Fig. 15 we compare the estimates based on the SPD multiplicity and the VZERO amplitude.The VZERO amplitude distribution is shown for two centrality classes selected by the SPD multiplicity.The distributions for the two centrality classes are reasonably well fitted with a Gaussian distribution.The resolution in the experimental definition of the centrality classes is evaluated event-byevent as the RMS of the distribution of the differences between the centrality determined over all estimators and the mean value of the centrality for the event.First the average value of the centrality c is calculated for each event by averaging the centrality determined by each estimator: c i is the centrality of an event determined by an estimator i, where i is the index running over all N = 6 centrality estimators used: VZERO (A and C), SPD, TPC, ZDC.In the next step, the centrality is weighted by ∆ i = c i − c : the difference between the centrality determined by each estimator and the mean value of the centrality from Eq. 8: This latter calculation is performed iteratively replacing c by the new value until convergence is achieved which typically occurs after the second iteration.Finally, the centrality resolution of an estimator is evaluated as the RMS of its ∆ i distribution for each centrality.
The ZDC-ZEM estimator is ignored for peripheral events ( c > 35%) since its results are reliable only for the most central collisions.The resolution is shown in Fig. 16 (left panel) as a function of the centrality percentile.The resolution depends on the rapidity coverage of the detector used.The best centrality resolution is achieved when combining the VZERO-A and VZERO-C detector, due to the large pseudo-rapidity coverage (4.3 units in total).It ranges from 0.5% in central to 2% in peripheral collisions.The resolution obtained with the SPD and the TPC ranges from 1% in central to 3% in peripheral collisions ( 80%).
We measured the pseudo-rapidity dependence of the charged particle multiplicity at midrapidity [45] with the SPD, and at forward rapidity [46] using all the rapidity coverage of the SPD, the VZERO and the FMD detectors.The total charged particle multiplicity N ch is obtained by integration.The centrality resolution was scaled by √ N ch measured in the rapidity window of each detector (see right panel of Fig. 16).The figure shows that all the results are consistent on an arbitrary unit scale, except for the ZDC-ZEM estimator which is better for central collisions because it uses information from two detectors.
The centrality resolution was tested with a full HIJING and GEANT detector simulation.In the HIJING simulations the true value of the event centrality (c true ) is known for every given event.After using GEANT one obtains the signals in VZERO, SPD, TPC for the given event and hence using these centrality estimators can calculate the value of the c for the given event with Eq.8.The real centrality resolution, given for the given event by the difference between the c true and the c calculated for each extimator, is consistent with the one calculated with data.

Summary
Heavy-ion collisions can be characterized by the number of charged particles produced in the collision.In principle, when normalized to the trigger efficiency used to collect the data sample, the charged particle multiplicity could provide a measurement of the hadronic cross section.However, at the LHC the large cross section for EM processes contaminates the very peripheral collisions.This problem was overcome in two ways.
In the first method, dedicated simulations of hadronic and EM processes (Fig. 8) were performed and data were corrected for efficiency of the event selection (Fig. 7) and purity of the event sample (Fig. 9).In the second method, the measured multiplicity distribution was fitted with a Glauber calculation (Fig. 10).Both methods allow to determine a centrality value above which the background contamination is negligible and the event selection is fully efficient.The corresponding value of multiplicity and centrality is defined as the anchor point, and is used for the centrality normalization (Table 3).
Using the AP, the measured event sample can be divided in centrality classes which correspond to well defined percentiles of the hadronic cross section.Several approaches were developed.The first method uses charged particle multiplicity (measured by various detectors, with different rapidity coverage, such as the VZERO, the SPD, and the TPC).The second method uses the ZDC, which measures the nucleon spectators directly, as well as the correlation to the ZEM energy in order to resolve the ambiguity due to nuclear fragmentation.The centrality is obtained from linear functions that fit the contours of the classes defined by the VZERO, in the ZDC-ZEM plane (Fig. 13).As standard method, typically used in ALICE physics analyses, we used the NBD-Glauber fit to the VZERO amplitude (Fig. 10) to determine the AP, and the other methods described to asses a systematic uncertainty on the centrality determination.
The resolution of the centrality determination, which depends on the pseudo-rapidity coverage of the detector used, was determined as the weighed RMS of all the estimates; it ranges from 0.5% in central to 2% in peripheral collisions (Fig. 16).
Finally, mean numbers of the relevant geometrical quantities, such as N part and N coll , were calculated for typical centrality classes, using the Glauber Model and the fit to the measured multiplicity distribution (Table 1).This fit creates a mapping between a measured quantity and one obtained with a phenomenological calculation for which the geometrical properties are known.The results, nearly identical to those obtained for centrality classes defined by classifying the events according to their impact parameter, provide a general tool to compare ALICE measurements with those of other experiments, at different energies and with different colliding systems as well as theoretical calculations.

A Tables
As described in Sec. 2, for the physics in ALICE analyses the average values of N part , N coll , or T AA for centrality classes defined by sharp cuts in the impact parameter distributions are used.These are reported in Table 1.Therefore N part , N coll and T AA depend exclusively on the nuclear geometrical parameters, and not on any measured quantity.Their uncertainty is calculated by varying the parameters of the Glauber calculations (i.e. the parameters of the Woods-Saxon and the hadronic cross section σ inel NN ) by the known uncertainty.We label the N part calculated with this procedure as N geo part .Another possibility, discussed in Sec.5.2, is to define the average values of N part , N coll , or T AA for centrality classes by sharp cuts in the fitted multiplicity distribution.Following this strategy, it is also possible to incorporate in the uncertainty, besides the uncertainties related to the Glauber calculation, those related to the measurement of the AP: the experimental region which is actually being used for the physics analyses, because it is free of background and the trigger efficiency is known.In this case, the AP can be varied by the uncertainty that was estimated (90% ± 1%) and recalculate N part , N coll and T AA with these variations.The N part calculated with this procedure is labled as N data part .The variations for the AP are labelled N data+ part and N data− part respectively.

Fig. 2 :
Fig. 2: Geometric properties of Pb-Pb collisions at √ s NN = 2.76 TeV obtained from a Glauber Monte Carlo calculation: Impact parameter distribution (left), sliced for percentiles of the hadronic cross section, and distributions of the number of participants (right) for the corresponding centrality classes.

Fig. 3 :
Fig. 3: Sensitivity of N part (left) and N coll (right) to variations of parameters in the Glauber Monte Carlo model of Pb-Pb collisions at √ s NN = 2.76 TeV.The gray band represents the RMS of N part and N coll respectively.It is scaled by a factor 0.1 for visibility.

Fig. 4 :
Fig. 4: Time distribution of signals in the VZERO detector on the A side.The peaks corresponding to beam-beam, beam-gas and satellite collision events are clearly visible.

−
V0AND: signals in VZERO-A and VZERO-C; − 3-out-of-3: signals in VZERO-A and VZERO-C and at least 2 chips hit in the outer layer of the SPD; − 2-out-of-3: two of the three conditions listed above.

Fig. 6 :
Fig.6: (Color online) Correlation between signals in the two neutron zero-degree calorimeters, ZNA and ZNC.The figure is taken from[9].Single electromagnetic dissociation events produce signal in only one of the calorimeters.Mutual dissociation and hadronic interactions populate interior of the plot and can be distinguished from each other by the signal in ZEM.

Fig. 7 :
Fig. 7: (Color online) Efficiency of the three online triggers (2−out − of−3, V0AND, 3−out − of−3) used for Pb-Pb collisions as a function of the VZERO amplitude calculated with HIJING and AMPT, and measured in dedicated pp runs.The efficiency in the simulation has been calculated for events with N part = 2.

Fig. 8 :
Fig. 8: (Color online) VZERO amplitude distribution in data (red points) and simulations with the V0AND interaction trigger.The data are compared to the sum of HIJING + QED + STARLIGHT simulations (histogram) with the same event selection.

Fig. 9 :
Fig. 9: (Color online) Purity of the 3 online interaction triggers (2-out-of-3, V0AND, 3-out-of-3) and other event selections used for Pb-Pb collisions as a function of the VZERO amplitude calculated with HIJING, STARLIGHT and QED simulations.The dashed line indicates 90% of the hadronic cross section.

3 . 1 )
Figure10shows the distribution of VZERO amplitudes for all events triggered with the 3-out-of-3 trigger (see 3.2) after removing the beam background (see 3.3.1),part of the EM background with the ZDC cut (see 3.3.2 ) and a Z-vertex cut |z vtx | < 10 cm.The multiplicity distribution has the classical shape of a peak corresponding to most peripheral collisions (contaminated by EM background and by missing events due to the trigger inefficiency), a plateau of the intermediate region and an edge for the central collisions, which is sensitive to the intrinsic fluctuations of

Fig. 10 :
Fig. 10: (Color online) Distribution of the sum of amplitudes in the VZERO scintillators.The distribution is fitted with the NBD-Glauber fit (explained in the text) shown as a line.The centrality classes used in the analysis are indicated in the figure.The inset shows a zoom of the most peripheral region.

Fig. 11 :
Fig. 11: (Color online) Centrality dependence of dN ch /dη per participant pair as a function of N part , measured in the Pb-Pb data at √ s NN = 2.76 TeV fitted with various parametrizations of N part and N coll , calculated with the Glauber model.The fit parameters are given in the figure.Data are from [35].

Fig. 12 :
Fig. 12: (Color online) Distribution of the VZERO amplitude zoomed in the most peripheral region.The distribution is compared to the NBD-Glauber fit and to the sum of the HIJING + STARLIGHT + QED simulations.

Fig. 13 :
Fig.13: (Color online) Spectator energy deposited in the ZDC calorimeters as a function of ZEM amplitude.The same correlation is shown for different centrality classes (5%, 10%, 20% and 30%) obtained by selecting specific VZERO amplitudes.The lines are a fit to the boundaries of the centrality classes with linear functions, where only the slope is fitted and the offset point is fixed (see text).

Fig. 14 :
Fig.14:(Color online) VZERO amplitude distribution of events of various centrality classes selected from the correlation between ZDC and ZEM amplitudes explained in the text.

Fig. 15 :
Fig. 15: (Color online) Top: Correlation between SPD multiplicity and VZERO amplitude.The rapidity coverage of each detector is indicated on the figure.Bottom: VZERO amplitude distributions for the centrality classes selected by SPD.Two centrality classes (1-2% and 40-45%) are indicated and fitted with a Gaussian.

Fig. 16 :
Fig. 16: (Color online) Left: Centrality resolution ∆ i for all the estimators evaluated in the analysis.Right: Resolution, in arbitrary units, scaled by √ N ch measured in each detector.
T HR  ch) or an energy deposited in the ZDC below a given value (E T HR ZDC ) in the ZDC energy distribution dσ /dE ZDC

Table 1 :
Geometric properties (N part , N coll , T AA ) of Pb-Pb collisions for centrality classes defined by sharp cuts in the impact parameter b (in fm).The mean values, the RMS, and the systematic uncertainties are obtained with a Glauber Monte Carlo calculation.

Table 2 :
Parameters of the fit to the charged particle multiplicity for the three different parametrizations discussed in the text, with error and χ 2 /NDF.

Table 3 :
Comparison of the percentage of the hadronic cross section above the VZERO amplitude chosen as AP (V0 AP ) for various cases considered in the systematic studies of the Glauber fits and with HIJING and AMPT simulations.

Table A .
1: N part for Pb-Pb collisions at √ s NN = 2.76 TeV with the corresponding uncertainties derived from a Glauber calculation.The N data part are calculated from the NBD-Glauber fit to the VZERO amplitude, while the N geo part are obtained by slicing the impact parameter distribution.N data part is also calculated for two variations of the AP, i.e. moving it to 91% ( N data+ Same comparison is done in Tables A.2 and A.3 for N coll and T AA respectively.TableA.4gives the comparison for the three quantities but for bigger centrality classes.

Table A . 2 :
Same as Table A.1 for N coll .

Table A . 3 :
Same as Table A.1 for T AA .

Table A . 4 :
Same as above with bigger centrality classes